Adoption of artificial intelligence in medical imaging results in faster diagnoses and reduced errors, when compared to traditional analysis of images produced by X-rays and MRIs. AI brings more capabilities to the majority of diagnostics, including cancer screening and chest CT exams aimed at detecting COVID-19. Both tech giants and growing startups are putting their efforts into releasing AI-based solutions. What opportunities and challenges do AI techniques bring to the table?

The Power of AI, ML, and DL

The diagnostic imaging market generally includes MRI systems, CT scanners, X-ray systems, ultrasound imaging systems, nuclear imaging systems, and mammography systems. These medical imaging systems capture internal images of a human body, and the role of Artificial Intelligence is to improve both the speed and accuracy of the system and standardize the process of diagnosis as well. 

Broadly speaking, Artificial Intelligence makes a solution smart enough to perform tasks like human doctors—detect abnormalities, segment anatomies, and measure lesions. Often used in AI, Machine Learning algorithms can extract and generalize hidden patterns and abnormalities that are sometimes difficult for a diagnostician to find. As a subset of ML, Deep Learning algorithms are trained on deep neural networks to classify objects in images. 

For example, researchers collect data on lesions in eye tissue and spinal cords from thousands of patients, then feed the algorithms with this information. Thus, AI/ML/DL-based healthcare industry digital solutions can offer hospitals and diagnostic centers more accurate and efficient diagnostic decisions based on solid data. 

Over the past few years, the adoption of AI techniques in the healthcare industry has gained a considerable amount of traction. According to Transparency Market Research, the global AI in medical imaging market was worth the US $384.7 million in 2019 and is expected to reach US $7.4 billion by the end of 2027 at a CAGR of 45%. 

How AI is Transforming Major Medical Imaging Systems

Computed Tomography (CT)

Computed tomography helps to identify many severe diseases, including internal brain hemorrhaging, kidney or bladder stones, and tumors. Typical diagnostic scanners create detailed 2D and 3D images of organs, bones, and soft tissue. When imaging, a CT scan uses x-rays, a form of ionizing radiation, which is generally known to increase the risk of cancer. 

AI (and machine learning in particular) has the potential to reduce the amount of radiation exposure and improve reconstruction of CT images—alongside its proven capability automating the workflow, eliminating errors, and increasing workload. 

Since the COVID-19 outbreak, the number of CT chest screening exams has grown exponentially. A recent study has demonstrated that AI using the deep learning method is now able to distinguish the virus from community acquired pneumonia. 

X-rays 

X-ray systems are widely used to produce black and white images of a human body to detect dental decay, tumors, and arthritis. While CT scanning is often criticized for its ionizing radiation, X-rays, with a lower level of exposure from electromagnetic waves, have their own drawbacks—they have limited accuracy in chest tests, that result in missed lesions, including early-stage lung cancer. In these cases, Artificial Intelligence can not only speed up testing but also improve image interpretation. For example, to increase the capabilities of X-rays, Google developed deep learning models that help classify clinically important findings on frontal chest radiographs. 

Magnetic Resonance Imaging (MRI) 

Magnetic resonance imaging uses a strong magnetic field and radio waves to test human organs and soft tissue from head to toe—with no radiation exposure. MRI scans, which produce high-resolution 3D images from multiple 2D images, work best to diagnose anomalies, disorders, and injuries of the brain and spinal cord. 

Today, Artificial Intelligence enhances MRI scans by automatically post-processing the imaging datasets. One of the outstanding examples in the market is the AI-powered, cloud-based AI-Rad Companion from SIEMENS Healthineers. Thanks to deep learning algorithms, the solution assists radiologists—automatically detecting and highlighting abnormalities, measuring lesions, segmenting anatomies, clarifying the location, creating a deviation map, and generating a report. At the same time, the software lets specialists manually mark and characterize lesions. 

Furthermore, Artificial Intelligence techniques are continually reducing the time a patient needs to spend inside the magnet without lowering image quality. 

Ultrasound Systems

Ultrasound imaging systems use sound waves to capture images of internal organs to diagnose lesions in livers, hearts, and kidneys and monitor the health of babies in the womb. Studies show that Ultrasound imagery with AI capabilities has the potential to diagnose arrhythmia more precisely compared to traditional ECG exams. 

Optical Coherence Tomography (OCT)

Optical coherence tomography helps to detect glaucoma, diabetic retinopathy, and retina diseases by taking pictures of different eye tissue layers. Through incorporating deep learning into OCT, diagnosticians have gotten a more accurate way to detect severe eye diseases like diabetic retinopathy. 

In a recent study, researchers focused on imaging the back of the eye. Artificial Intelligence methods helped them extract two main tissue layers, the retina and choroid, and then the researchers fed the algorithms with new data on changes in eye tissue caused by refractive errors, disease, and aging. 

What About Cancer Diagnostics? 

Along with dozens of anomalies, disorders, and injuries, digital imaging in medicine identifies changes in the human body caused by cancer. MRIs and CT scans are useful for non-invasive diagnosis of many types of cancer (breast and lung, in particular), detecting the precise location, stage, effectiveness of treatment, and in some cases helping to decide whether a biopsy is necessary. Of course, imaging tests have their limitations—for example, they can find large groups of cancerous cells but will miss single cells. 

With the power of ML and DL algorithms, diagnostic tools can analyze medical images faster and more accurately than ever before. This is why many tech giants and growing startups put their efforts into AI-based solutions for diagnosing cancer. For example, Softeq helped Veriskin launch a custom skin cancer screening device. Follow-up studies demonstrated the efficiency of the device for determining cancerous lesions: the gadget detected malignant skin growth without the need for extra biopsies or examinations.  

The Benefits of Artificial Intelligence in Medical Imaging 

The most common problems that Artificial Intelligence solves in the healthcare industry generally involve speed, accuracy, and workload. AI capabilities enhance the internet of things in medicine—not only IoT-based tools that produce pictures of human bodies, but also hospital automation systems, telehealth solutions, and medication trackers. Here are potential benefits AI techniques bring to medical imaging (both for diagnosticians and patients):  

Automation. AI brings higher automation to the workflow—automated registration of images, segmentation of anatomies, measurement of lesions. It also provides assistance in the interpretation of cases. In some cases, technicians having a degree in radiology is becoming an option, not a must.

Higher Accuracy. Artificial Intelligence provides more accuracy in diagnostics with expanded image datasets feeding algorithms, which help to detect cancerous cells or lesions in eye tissue. 

Cost. Adoption of AI reduces the cost of medical imaging tools and lowers the price of diagnostic procedures, which means more patients around the world have the opportunity to be tested. 

Productivity. AI-assisted software, which is able to analyze and interpret images faster than humans, increases doctor productivity; AI-based solutions speed up image processing in healthcare. 

Image Quality. An overall increase in the quality of images is achieved through denoising and reconstructing. 

Workload. Reduced workload shortens waiting lists and multiplies the number of tests that can be conducted. 

Speed. AI methods reduce screening time when patients are in CT or MRI machines.  

Radiation Exposure Level. AI techniques help to lower radiation exposure for patients during CT screening and X-ray exams—as it can accurately reconstruct high-quality images from low-quality originals. 

Non-Invasive Diagnostics. Faster diagnostics using non-invasive methods can result in better outcomes for patients with cancer detected at early stages. 

Top 3 Things to Consider When Developing an AI-based Medical Imaging Solution

Compliance with Regulatory Standards 

This may include industry-specific standards and regulations, such as HIPAA, HL7, GMP, DICOM, and FDA requirements. While the HIPAA (Health Insurance Portability and Accountability Act) addresses the use and disclosure of individual health information and protects personal data, such as the patient’s name, address, and past medical history, DICOM refers to the integration of all medical imaging devices supporting the same file format. 

To incorporate all required capabilities, the solution may need additional development and budget for clinical trials, to demonstrate efficiency and safety. Certification can be a costly endeavor. At the same time, more guidance from regulators simplifies the certification path for novel medical devices. 

The biggest challenges companies may face developing medical imaging solutions is access to multiple datasets with health information, and accurate and legal use of this data. 

Core Functional Requirements that Determine the Tech Stack  

These requirements will define how your solution will be equipped on both the hardware and software levels. For example, you may need AI-assisted software that will be able to annotate images produced by target medical equipment—CT scans, MRI machines, or ultrasound systems. 

To deliver the solution, developers can integrate AI methods into a custom embedded, mobile, or web application, feed the data to a neural network and create an infrastructure to store, analyze, and process data in-house or in the cloud. Required tools and techniques may involve using machine learning platforms, such as TensorFlow, and coding with a specific programming language like Python. 

Releasing a medical device that requires tech decisions at the intersection of hardware, software, user experience and design is another story altogether. In this case, tech vendors will produce hardware (with a processor and various incorporated components, such as a display, LEDs, and audio speakers), create software with the relevant feature set and UI, and enable the hardware devices and software to work together. 

Prioritize User Experience and Provide Assistance 

Today, diagnosticians need skills to use AI-powered tools—to validate automated test results and recognise potential errors made by algorithms, as well as familiarity with the process of image analysis and interpretation. Engineers, in turn, are tasked with creating an experience based on the level of user expertise. This will help make the interaction intuitive and speed up the learning process. In some cases, you may consider providing guided tours or workshops. 

Bottom Line

Greater adoption of Artificial Intelligence in medical imaging empowers diagnosticians and improves the quality of life of their patients. Smart techniques have enhanced medical imaging systems, helping detect abnormalities and measure lesions better, faster, and cheaper. 

The increasing amount of research in AI brings opportunities to tech companies for developing new products or services and expanding to new markets. The opportunities come with new challenges, which force techies to extend their competencies and upgrade their level of expertise. Moreover, it is hard to imagine the release of an AI-based medical device without access to datasets with health information, compliance with industry-specific regulations, a budget for clinical trials, and a focus on security from day one.